Dynamic Power Management for Long-Term Energy Neutral Operation of Solar Energy Harvesting Systems Bernhard Buchli, Felix Sutton, Jan Beutel, Lothar Thiele Computer Engineering and Networks Laboratory Swiss Federal Institute of Technology (ETH) Zurich Zurich, Switzerland
{bbuchli, fsutton, beutel, thiele}@tik.ee.ethz.ch
Abstract
1
In this work we consider a real-world environmental monitoring scenario that requires uninterrupted system operation over time periods on the order of multiple years. To achieve this goal, we propose a novel approach to dynamically adjust the system’s performance level such that energy neutral operation, and thus long-term uninterrupted operation can be achieved. We first consider the annual dynamics of the energy source to design an appropriate power subsystem (i.e., solar panel size and energy store capacity), and then dynamically compute the long-term sustainable performance level at runtime. We show through trace-driven simulations using eleven years of real-world data that our approach outperforms existing predictive, e.g., EWMA, WCMA, and reactive, e.g., ENO-MAX, approaches in terms of average performance level by up to 177%, while reducing duty-cycle variance by up to three orders of magnitude. We further demonstrate the benefits of the dynamic power management scheme using a wireless sensor system deployed for environmental monitoring in a remote, high-alpine environment as a case study. A performance evaluation over two years reveals that the dynamic power management scheme achieves a two-fold improvement in system utility when compared to only applying appropriate capacity planning.
The performance level achievable by an embedded system is ultimately limited by the energy available to operate the device. Due to the predominantly remote deployment locations of Wireless Sensor Networks (WSNs) and the lack of dependable power sources, the motes comprising these networks generally rely on batteries for delivering the energy to fulfill their intended task. However, due to the finite capacity of the energy storage element, i.e., battery, the motes are highly energy constrained and suffer from a severely constrained lifetime. To improve the system’s achievable performance level and extend the lifetime, ambient energy harvesting, particularly in the form of solar energy harvesting, has been established as a feasible alternative to purely battery powered devices in outdoor WSN applications [20]. Using a real-world application scenario [3], which requires high system availability, and relies on sensing technology characterized by high energy demands, we investigate if the two conflicting goals, i.e., high system performance and lifetime on the order of multiple years, can be simultaneously achieved with solar energy harvesting systems. A broad range of application scenarios, e.g., [9,11,22,24], benefit from a minimum supported performance level that can be sustained over time periods on the order of multiple years. A system enhanced with energy harvesting capabilities can – in theory – operate indefinitely as the energy store can be replenished periodically. Experience has shown, however, that enhancing a battery operated device with energy harvesting capabilities will by itself neither provide a lower bound on the expected sustainable performance level, nor guarantee uninterrupted long-term operation [21]. The reason for this is the dependence on an uncontrollable energy source [15], i.e., the sun, which exhibits high short-term fluctuations due to meteorological conditions that are hard to model [8] and difficult to predict [13]. Contemporary power management techniques deal with the highly variable energy harvesting opportunities by dynamically adapting the system’s performance level at runtime such that Energy Neutral Operation (ENO) [15] may be achieved. Informally, a system is said to operate in an energy neutral mode if the energy consumed over a given time period δ is less than or equal to the energy harvested during the same time period. Due to practical limitations, ENO is generally interpreted such that the battery fill-level B f ill at
Categories and Subject Descriptors C.4 [Performance of Systems]: Design studies, modeling techniques
General Terms Algorithms, design, experimentation * Keywords Energy neutral operation, solar energy harvesting, wireless sensor networks
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from
[email protected]. SenSys’14, November 3–6, 2014, Memphis, TN, USA. Copyright 2014 ACM 978-1-4503-3143-2/14/11 ...$15.00 http://dx.doi.org/10.1145/2668332.2668333
Introduction
the end of period δ must be greater than or equal to that at the beginning, i.e., B f ill (t + δ) ≥ B f ill (t) [15, 18, 23]. Given ENO as the fundamental bound of energy harvesting systems, numerous methods that attempt to achieve this objective have been proposed, e.g., [15–18,23]. These can be classified as (i) predictive, and (ii) reactive approaches. Predictive approaches, e.g., [15, 18], attempt to satisfy ENO by predicting the harvestable energy during a future time slot, and adapt the performance level accordingly. However, predicting future meteorological conditions is highly complex and may be computationally prohibitive [8]. Therefore, acceptable prediction accuracy with the limited computational resources available on contemporary motes has so far only been possible for short prediction windows, i.e., δ on the order of minutes to hours. Reactive approaches, on the other hand, attempt to satisfy energy neutrality by scheduling the performance level in response to changes in the source. This can be done by measuring the energy generation directly, or, as is commonly done, through monitoring the battery fill-level [23], or super-capacitor voltage [16]. The performance of a storagereactive approach strongly relies on the accuracy of the battery State-of-Charge indication. Current implementations of the above two classes adapt the system duty-cycle in response to, or expectation of, shortterm variations of the energy source, and thus tend to suffer from high duty-cycle variance. Duty-cycle variance is an important consideration, e.g., for surveillance applications, where the system should be available with equal probability at any given point in time [12]. In this work we turn our attention to enabling long-term energy neutral operation for solar energy harvesting systems. Rather than predicting or reacting to the source’s short-term variations, we argue that the source’s long-term dynamics must be considered both for dimensioning the power subsystem and devising the dynamic energy management scheme. We leverage the approach discussed in [7] to provision the power subsystem, i.e., battery and solar panel, such that short-term fluctuations can be absorbed. We further devise a long-term energy-predictive dynamic power management technique that can compute the long-term sustainable performance level at runtime. The contributions of this work are as follows. First, we present an end-to-end solution for enabling Long-Term Energy Neutral Operation (LT-ENO) for solar energy harvesting systems. Our approach encompasses (i) a power subsystem capacity planning algorithm based on an astronomical solar radiation model, and (ii) a dynamic energy management scheme, which is based on the same astronomical model, and that can enable uninterrupted operation with very low duty-cycle variance. Second, through simulation with eleven years of data at three different geographical locations, we show that our algorithm outperforms the State-of-theArt in energy-predictive [15, 18], and battery-reactive [23] performance scaling approaches in terms of average sustainable performance level by up to 177%, energy efficiency by up to 184%, and duty-cycle stability by up to three orders of magnitude, while incurring zero downtime, i.e., system availability of 100%. Finally, we exemplify the benefits of
our approach using an X-S ENSE environmental monitoring system [3] deployed over two years in a high-alpine environment, and demonstrate that significant improvements in system utility can be achieved without risking downtime due to power outages. The rest of this paper is structured as follows. Sec. 2 reviews the State-of-the-Art in power subsystem capacity planning, energy prediction schemes, and harvesting aware dynamic performance scaling techniques. Sec. 3 reviews the power subsystem capacity planning approach. The dynamic power management technique for LT-ENO is discussed in detail in Sec. 4. In Sec. 5 the proposed technique is evaluated through simulation with eleven years of data for three different locations, while Sec. 6 provides performance results obtained with a system deployed over a period of two years. Sec. 7 concludes this work with a summary of key findings.
2
Related Work
Capacity Planning. The importance of proper capacity planning for solar energy harvesting systems has been introduced in [15], and a systematic technique was proposed. The approach relies on the availability of a representative energy generation profile and the known system consumption to compute the battery capacity. The limitations of this approach are two-fold. First, the input trace, i.e., energy profile, must be representative of the conditions at the intended deployment site, and cover at least one full annual solar cycle to yield a suitable battery capacity. Second, the panel size is not considered a design parameter, thus preventing the designer from optimizing the power subsystem with respect to cost, physical form-factor, etc. A recently proposed approach [7] mitigates the aforementioned shortcomings. The authors propose a capacity planning algorithm that relies on an astronomical model to approximate the energy profile at the intended deployment site. With this approach, the designer can vary all important design parameters to obtain the specifications of a suitable power subsystem for a given application without the need for extensive trace data. Through simulation with ten years of trace data, it was shown that the power subsystem obtained enables uninterrupted operation if the actual total energy generation is at least 80% of the modeled expectations. In this work we develop a dynamic power management scheme that builds upon this capacity planning technique. Energy Management. In the seminal work on energy harvesting theory [15], the first dynamic duty-cycling scheme for solar energy harvesting systems was proposed within a theoretical framework that defines Energy Neutral Operation (ENO) as the fundamental limit of energy harvesting systems. ENO is achieved if the system never consumes more energy than what it can harvest over a given time period δ, i.e., the battery fill-level B f ill (t + δ) is greater than or equal to B f ill (t). With their approach, a day is discretized into slots of equal duration δ, and the expected energy input for each slot is learned with an Exponentially Weighted Moving Average (EWMA) filter. Each slot’s respective duty-cycle is then computed by considering the mismatch between expected and actual energy input. However, due to limited correlation between past and future weather conditions, this approach
Offline Long-Term Capacity Planning (LT-CP) [7] Deployment Parameters: Latitude, Orientation, Inclination
Runtime Long-Term Dynamic Power Management (LT-DPM)
Environmental Parameter: Ω
Technology Parameters: A pv , Ppv , η pv , ηout , Pcc , ηcc
Harvesting Conditioned Energy Availability Model Ein Sec. 3.2
System Parameters: Psys , DCsys
Power Subsystem Dimensioning Sec. 3.3
Measured Harvested Energy Ereal (Sec. 4.2)
Sec. 4 Energy Availability Model Adjustment α, Ebin
Nominal Battery Capacity Bnom
Duty-Cycle Adaption DC
Figure 1: Process flow for long-term solar energy harvesting capacity planning and dynamic power management. Dashed boxes and arrows represent user inputs. The offline capacity planning algorithm computes the achievable duty-cycle and required battery capacity for the given input parameter set, and the dynamic power management algorithm adjusts the system performance level at runtime according to the observed conditions.
achieves acceptable prediction accuracy only for prediction windows on the order of hours. Weather Conditioned Moving Average (WCMA), proposed in [18], improves upon EWMA’s prediction accuracy. The authors not only consider the harvested energy in the same time slot during previous days, but also incorporate current weather conditions to obtain the expected energy input in the current slot. While achieving an almost three-fold improvement in prediction accuracy over EWMA, it is not clear if and how this improvement translates into increased system performance and/or energy neutrality. This approach is also constrained by short prediction windows. More recently, the use of professional weather forecast services have been considered to predict the disposable energy [19]. The authors formulate a model to translate weather forecasts into solar or wind energy harvesting predictions. While it is unclear what baseline is used, the authors conclude that their energy predictions are more accurate than those based on past local observations. In [16] and [23], model-free approaches to dynamic performance scaling are presented. In [23], a technique from adaptive control theory, i.e., Linear-Quadratic Tracking, is used to dynamically adapt the system’s duty-cycle based on the battery State-of-Charge and so ensure ENO. For the datasets evaluated, the authors report between 6 and 32% improvement in mean duty-cycle, and between 6 and 69% reduction in duty-cycle variance when compared to EWMA. Similarly, in [16] a Proportional-Integral-Derivative (PID) controller monitors the energy storage element, and the dutycycle is adapted such that an expected voltage level of the storage element (a super-capacitor in this case) is maintained. While presenting low-complexity solutions, both of these approaches suffer from high duty-cycle variability, and rely on a well performing battery State-of-Charge approximation algorithm. The PID approach additionally requires parameter tuning, for which solutions exist in the literature.
3
Capacity Planning for Long-Term Energy Neutral Operation
Rather than modeling the energy source’s highly variable short-term dynamics and adjust the performance level accordingly, we propose a long-term energy neutral power management scheme for solar energy harvesting systems. Our approach, illustrated in Figure 1, first invokes a designtime power subsystem capacity planning algorithm to deter-
mine the required battery capacity given a set of input parameters that characterize the system and its environment. The intricate trade-offs between battery capacity, and the system and environmental parameters are discussed in [7]. This algorithm uses an astronomical model to estimate the long-term energy availability based on the annual solar cycle. Then, at runtime, the proposed algorithm dynamically computes the performance level, i.e., duty-cycle, based on an adjusted energy availability model such that long-term energy neutrality can be sustained. The energy model and the capacity planning approach follow [7] and are briefly reviewed in this section. The novel dynamic power management scheme is discussed in detail in Sec. 4, and evaluated in Sec. 5 and 6.
3.1
System Architecture, Load Model, and System Utility
In this work we assume a harvest-store-use architecture, as described in [20], in which the energy to operate the system is always supplied by the battery. We further assume that the power Psys dissipated from the battery includes all consumers present in the system, e.g., power conditioning and other supervisory circuitry. Further considering that contemporary embedded systems can operate in sleep modes with ultra-low power dissipation, we ignore its contribution and define the total daily energy Eout (d) necessary to sustain a required performance level DCsys (d) on calendar day d as given in (1), where γ = 24 hours. Note that we ignore battery leakage here, but it can be integrated into the load model. Eout (d) = γ · DCsys (d) · Psys ,
∀d ∈ Z+
(1)
For now, we assume a one-to-one relationship between performance level DCsys (d) and utility of the system U, i.e., U(DCsys (d)) = DCsys (d) [9]. We revisit this topic in Sec. 6, where we refine the definition of system utility in the context of a real system. Note that we are not concerned with how the energy is scheduled and consumed over the course of the day, but rather provide information about disposable energy to an application specific task scheduler. Details on local scheduling of the available energy, and network-wide balancing of the energy budget by changing the communication and/or sensing patterns are beyond the scope of this paper, as they are highly application specific. For example, a scheduler’s primary focus may be planning the available energy such that a minimum level of operation may be sustained. Any excess energy may then be used to improve sensing,
.
3.5
Ein(d)
Eout(d)
Ereal(d)
Energy [Wh]
3 2.5 2 1.5 1 0.5 0
d
d
0
d2
1
90
180
270
360
90
Calendar Day
Figure 2: Solar energy profile for a particular geographical location and energy harvesting setup. Surplus energy generated by the panel is indicated with the hatched area; the energy deficit is shown by the cross-hatched area.
processing or communication.
3.2
Harvesting Conditioned Energy Availability Model
A crucial step in capacity planning consists of estimating the theoretically harvestable energy at a specific point in space and time. Figure 2 illustrates the amount of solar energy harvested at a particular geographical location and given harvesting configuration. The figure shows the total daily energy input Ereal (d) at the end of each calendar day d, and illustrates the high short-term (day-to-day) variability and long-term periodicity (year-to-year) of the source. Also shown is the modeled total expected harvestable energy Ein (d) on calendar day d such that true energy conditions are closely approximated, i.e., (2) holds where N is the number of days. N
3.3
d2
∑ (Eout (d) − Ein (d)) ≤ B
N
= ∑ Ein (d) ∼
∑ Ereal (d),
d=1
d=1
Power Subsystem Dimensioning
In this section we review the process of computing the power subsystem capacity using the energy availability model such that energy neutral operation over the source’s seasonal cycle, i.e., one year, can be achieved. At this point we assume a perfect battery, i.e., no inefficiencies. For a discussion including various battery inefficiencies, the reader is referred to [7]. For the purpose of power subsystem capacity planning we assume a constant daily energy demand Eout (d) that must be met. Note that we explicitly keep the dependence on calendar day d, since the energy consumption at runtime varies with the dynamically chosen daily duty-cycle (see Sec. 4). Referring to Figure 2, we observe that the intersections between the energy consumption Eout (d) and approximated energy input Ein (d) partition the annual solar cycle into time regions of energy surplus, i.e., Ein (d) > Eout (d) ∀d ∈ [d0 , d1 ), and energy deficit, i.e., Ein (d) < Eout (d) ∀d ∈ [d1 , d2 ). According to the model assumptions, the minimum battery capacity B required to support the system during periods of energy deficit is indicated with the cross-hatched area in Figure 2, and formally stated in (4). The first term on the left-hand side defines the amount of energy that is necessary to support the system operation, while the second term represents the expected energy input. The difference is then the minimum required battery capacity.
N >> 1
The method to compute Ein (d) is based on a simplified astronomical model to estimate the theoretical solar radiation Eastro (t, d, L, θ p , φ p , Ω). It is parameterized by the time t in hours of calendar day d, the intended deployment site’s latitude L, and the panel’s orientation and inclination angles φ p and θ p , respectively. Finally, the environmental parameter Ω represents the expected average meteorological conditions. This is the only unknown input parameter, and can be approximated as described in [7]. Although not absolutely necessary, the availability of solar maps or solar energy traces can improve the approximation of the parameter Ω. Since we are concerned with electrical, as opposed to solar energy, the output of Eastro (·) must be conditioned by the technology parameters in Figure 1. These specify the panel’s surface area A pv , conversion efficiency η pv , and self-consumption and efficiency factors for supervisory and power conditioning circuitry, e.g., battery charge controller efficiency ηcc , and consumption Pcc . The maximum rated power output of the panel Ppv is used to evaluate the maximum energy E pv generated during one hour. Then, with the above parameters specified, the total electrical energy that can be harvested on calendar day d is approximated with (3). 24
Ein (d) = A pv ηcc η pv ∑ min(E pv , Eastro (t, d, ...))
(3)
t=1
While the astronomical energy model Eastro (·) may yield any resolution t, for the purpose of long-term energy neutral operation discussed in this work, daily sums are sufficient.
(4)
d1
(2)
In order to achieve uninterrupted operation over multiple years, it is not sufficient to only provision the battery for the period of deficit. The panel must be able to generate enough energy to recharge the battery in addition to the energy required to sustain operation during periods of energy surplus, i.e., d ∈ [d0 , d1 ). The constraint on energy generation by the panel is given in (5). d1
∑ (Ein (d) − Eout (d)) ≥ B
(5)
d0
The required battery capacity B can then be obtained by varying the performance level (i.e., DCsys (d)) and/or the panel area A pv and finding the intersections d0 , d1 , and d2 between Ein (d) and Eout (d) such that (4) and (5) hold.
4
Dynamic Power Management for LongTerm Energy Neutral Operation
In the previous section we described the design-time energy availability model and power subsystem capacity planning based on the long-term characteristics of the energy source. Assuming that the design-time model reflects the conditions at the deployment location to within some bounds, the system will be able to run at the performance level for which the power subsystem was designed [7]. However, in practice significant deviations from the model must be expected. Such deviations may be caused by transient phenomena, e.g., snow cover and foliage, or persistent occlusions due to trees and buildings. In this section we propose a dynamic power management scheme that can adapt
to deviations from the modeled assumptions by dynamically scaling the system performance level, and by doing so enable Long-Term Energy Neutral Operation (LT-ENO).
4.1
Dynamic Performance Scaling
As discussed in Sec. 3.3, in order to achieve long-term energy neutrality, the two constraints from (4) and (5) must be satisfied. The constraint in (4) states that the battery must be able to supply the difference in energy consumption and generation during periods of energy deficit, i.e., d ∈ [d1 , d2 ) (as shown in Figure 2). The second constraint states that, in order to ensure that the battery can be fully recharged during periods of energy surplus (d ∈ [d0 , d1 )), the panel must generate energy in excess of what is required to sustain shortterm operation. To satisfy these two constraints, we leverage the offline energy model to determine the sustainable system performance level. To exemplify our approach we consider a concrete example as illustrated in Figure 3. Without loss of generality, we assume that the design-time model Ein (d), which was used to obtain the battery capacity B given panel size A pv , overestimates the actual energy conditions Ereal (d). For simplicity we ignore battery inefficiencies in this discussion, but note that Algorithm 1 and the evaluation in Sec. 5 account for these effects. In the following we consider the end of day d and wish to compute the duty-cycle for the entire day d + 1 such that long-term energy neutrality may be achieved. To react to deviations from the modeled energy expectation, we first need to adjust the design-time energy model Ein (d) at runtime according to observed conditions. For this purpose, we define the model adjustment factor α in (6) to scale Ein (d), i.e., Ebin (d) = αEin (d), ∀d. The adjustment factor depends on the history window size W in days, which is used to tune the duty-cycle stability. The choice of W has a direct impact on the system’s responsiveness to variations in the energy profile, and therefore imposes a system tradeoff between duty-cycle stability and achievable performance level. The effects of the choice of the history window size W are discussed in Sec. 5.3.3. α=
∑dd−W Ereal (d) , ∑dd−W Ein (d)
0
